[MA] Design, Implementation, and Evaluation of an Intelligent Agent for Semantic Search in Structured E-Invoices

Start: 01.09.2025

End: 18.03.2026

Type: Master Thesis

Student: Alexander Hense

Supervisor: Prof. Dr. Freimut Bodendorf, Sebastian Schmid

Abstract: This thesis presents an intelligent invoice agent that uses natural language to query large invoice archives stored in Google Drive, powered by a ReAct architecture with LLM integration. The system parses XRechnung XML invoices, extracts structured data, and answers complex queries, from finding specific invoices to aggregating financial metrics across time periods and suppliers. It includes date expression normalization (quarters, months), caching, and metadata-based pre-filtering to reduce processing overhead. By combining LLM reasoning with deterministic filtering and aggregation, the agent enables business users to explore invoice data through conversational queries, transforming traditional document management into an interactive analytics platform.